DRAGENwiki · Article
Discontinuous Measurement: Recording a Sample of Behavior
Customize your experience
Sources checked. Clinical review pending.
Discontinuous measurement records a sample of behavior instead of every occurrence. The observation is split into equal intervals, and each interval is marked or left blank by one fixed rule.
Discontinuous measurement in 60 seconds
- A sample, on purpose. Discontinuous systems "capture only a sample of behavior during an observation." [3]
- Three common rules. Partial-interval recording, whole-interval recording and momentary time sampling are the three named in the technician outline. [2]
- The result is an estimate. Partial-interval recording "generates an estimate of the frequency and duration of behavior." [4] It is not a count.
- Each rule leans a known way. Partial-interval tends to run high, whole-interval tends to run low, and momentary time sampling has no set direction of error. [3]
- Choose by goal. Building a skill and reducing a behavior call for different rules. Shorter intervals and longer observations usually mean less error. [3]
Jump to: What makes it discontinuous · Three ways to score · Which way each leans · Choosing by goal · Interval and observation length · What a percentage can hide · Measuring yes and no · Example · Myths · Words to know · Need a PDF? · More from DRAGEN · Sources
What makes it discontinuous
The BCBA test content outline asks behavior analysts to "Distinguish between continuous and discontinuous measurement procedures" [1] and to "Design and apply discontinuous measurement procedures (e.g., interval recording, time sampling)." [1] The RBT outline asks technicians to "Implement discontinuous measurement procedures (e.g., partial & whole interval, momentary time sampling)." [2]
The difference is coverage. A continuous record catches every occurrence. A discontinuous record catches a sample, and it does so on purpose.
Fiske and Delmolino (2012) describe discontinuous systems as those that "capture only a sample of behavior during an observation." [3] These systems "involve dividing an observation into equal duration intervals, and scoring the occurrence or nonoccurrence of behavior within each interval." [3]
LeBlanc and colleagues (2016) explain why: "partial-interval recording is a discontinuous measurement procedure because some behaviors are deliberately not recorded." [4] As a result, it "does not produce complete data about any behavioral dimension but instead generates an estimate of the frequency and duration of behavior." [4]
The key word is estimate. The result is a percent of intervals. It points toward how often or how long the behavior happened, but it is not a count and not a timed duration.
It is common. Fiske and Delmolino report that Mudford, Taylor and Martin (2009) "found that 45% of studies published in the Journal of Applied Behavior Analysis from 1995 to 2005 used discontinuous measures of data collection." [3] Mudford and colleagues' own paper reports that 93 of the 168 articles reporting data on free-operant human behavior (55%) used continuous recording. [5]
Fiske and Delmolino add that research often measures more rigorously than everyday practice, so "this number is likely an underestimate of the extent to which discontinuous data collection methods are used in clinical and educational settings." [3]
Three ways to score an interval
All three start the same way. Pick an interval length, such as 10 seconds. Use a timer or a recorded beep to mark each interval. Then apply one scoring rule, the same rule every time.
Partial-interval recording (PIR). Mark the interval if the behavior happens at any point in it. In Fiske and Delmolino's words, "an occurrence is defined as an instance of target behavior that occurs at any time during the interval." [3]
Once an interval is marked, the observer can stop watching for the rest of it. LeBlanc and colleagues write that because partial-interval recording "does not require constant vigilance (i.e., one no longer needs to observe within an interval after a behavior has been scored), it is often used to measure high-rate behavior, as well as multiple forms of behavior." [4]
Whole-interval recording (WIR). Mark the interval only if the behavior lasts the entire interval. If the behavior stops for part of the interval, the interval stays blank. [3]
Momentary time sampling (MTS). Look only at the moment each interval ends. MTS "defines an occurrence only if the target behavior occurs as the interval ends, frequently in the last second of the interval." [3] Because it "does not require constant vigilance, it is often used to measure multiple forms of behavior or the behavior of multiple individuals." [4]
One interval, three answers. Picture one 10-second interval. A student writes for the first 4 seconds, then looks out the window for the last 6. (This example is made up.)
- Partial-interval: marked, because writing happened.
- Whole-interval: blank, because writing did not last all 10 seconds.
- Momentary: blank, because the student was not writing at the 10-second mark.
Reporting. For all three, "Data are reported as the percentage of intervals during which behavior was scored." [3] Ten marked intervals out of 60 is about 17%.
Which way each one leans
The three rules make different mistakes. Knowing which way each one leans is most of the skill.
- Partial-interval tends to run high. "PIR will consistently overestimate the true occurrence of behavior, since both a 1-s response and a 9-s response in each 10-s interval are coded identically as an occurrence." [3]
- Whole-interval tends to run low. "WIR will consistently underestimate the true occurrence of a behavior, with both a 1-s response and a 9-s response coded identically as a nonoccurrence." [3]
- Momentary time sampling has no set direction. "MTS is not associated with a characteristic direction of error." [3] LeBlanc and colleagues add that its error "is inconsistent (unlike the overestimation of partial-interval recording) and is exaggerated by long intervals and observation periods." [4]
How big can the error get? Fiske and Delmolino give an extreme case. Suppose that after treatment, a child is on task for only 3 seconds of every 10-second interval. Partial-interval data would show 100% on task when the true value is 30%. The graph would "falsely indicate that treatment was highly effective." [3]
Older studies point the same way. Meany-Daboul and colleagues (2007) summarize a 1975 comparison against continuous duration: "PIR consistently overestimated duration, whereas MTS either over- or underestimated duration; however, the margin of error associated with MTS was much smaller than that associated with PIR." [6]
Why the errors happen. Meany-Daboul and colleagues also report a 1991 analysis: error from both methods "is a result of the proportion of mixed intervals (i.e., those in which behavior occurs during only a portion of the interval)." [6]
An interval that the behavior fills completely, or not at all, comes out right under every rule. The trouble lives in the partly filled ones.
Tested by computer. In Wirth, Slaven and Taylor (2014), "A computer program simulated the application of momentary time sampling, partial-interval recording, and whole-interval recording methods on target events randomly distributed across an observation period." [7] Their full error tables are in the article's Supporting Information. [7]
Choosing by goal: building a skill or reducing a behavior
Fiske and Delmolino frame the choice around two kinds of work: "behavioral deficits (i.e., increasing low-occurrence behavior) and behavioral excesses (i.e., decreasing high-occurrence behavior)." [3] In plain words: a skill to build, or a behavior to reduce.
Building a skill.
- Their pick is momentary time sampling. "MTS has generally been shown to be the most accurate measurement system and is recommended for measurement in cases of treating behavioral deficits." [3]
- One caution. "MTS may have difficulty capturing low-frequency behavior" [3] at the start, and it may miss small early gains. [3]
- Partial-interval can hide growth. It inflates the starting level. For a skill that needs to happen at very high levels, such as staying on task, they do not recommend it. [3]
- Whole-interval is overly strict. It runs low, so the skill can look lower than it really is. They call it "an appropriate, but overly stringent measurement system in many situations." [3]
Reducing a behavior.
- Whole-interval can fake a drop. It is likely to show strong treatment effects even when the behavior continued. A behavior that lasts 9 seconds of every 10-second interval would be scored as never happening. "For this reason, WIR is not recommended in cases of treating behavioral excesses." [3]
- Partial-interval is strict. It runs high, so small decreases may not show at first. Still, "PIR data collection is appropriate for treatment of behavioral excesses, but will be very stringent." [3]
- When the goal is zero. When a behavior must be brought to zero, they note that only partial-interval recording could capture that accurately. If zero is not essential, they consider momentary time sampling appropriate. [3]
What other authors add.
- Continuous first. LeBlanc and colleagues write that when every event can be captured, "continuous measurement procedures are strongly preferred because they eliminate the error associated with estimate-based discontinuous measurement." [4]
- No whole-interval. In their words, "whole-interval recording is not recommended in our model." [4]
- Busy settings. When a person cannot watch for every occurrence, for example a teacher with 25 students, "a discontinuous measurement procedure in the form of momentary time sampling would likely be a more practical and preferred option." [4]
- Duration or frequency. Meany-Daboul and colleagues concluded that "the recommended discontinuous measurement method for estimating target responses appropriately measured using duration is MTS, and the recommended discontinuous measurement method for estimating the frequency of responses is PIR." [6] Their conclusion rests on data sets from five earlier treatment analyses, scored with 10-second intervals, and they note that treatments aimed at increasing behavior might give different findings. [6]
The one-line test. The BCBA outline sums up the job: "Select a measurement procedure to obtain representative data that accounts for the critical dimension of the behavior and environmental constraints." [1]
A quick decision path.
- Can you catch every occurrence? Record continuously. [4]
- Building a skill? Start with momentary time sampling. [3]
- Reducing a behavior? Partial-interval recording is the strict choice. Avoid whole-interval recording. [3] [4]
- Watching many learners or behaviors? Momentary time sampling is often the practical option. [4]
How long to make the intervals and the observation
Shorter intervals mean less systematic error. In Fiske and Delmolino's words, "the briefer the interval duration, the less systematic error will be introduced into a measurement system, regardless of whether PIR, WIR, or MTS are implemented." [3]
Shorter intervals also cost effort. Brief intervals ask more of the observer, which can add human error. [3] In one study they describe (Hanley and colleagues, 2007), observers agreed well on preschoolers' engagement with momentary time sampling intervals as brief as 3 seconds, but agreement fell to unacceptable levels below 3 seconds. [3] As intervals approach 1 second, the method is close to continuous recording anyway. [3]
Ten seconds is common, not magic. "The most common interval in the available literature implementing these systems is 10 s, but in most cases 10 s intervals with PIR will still overestimate behavior." [3] Momentary time sampling holds up better: "MTS tends to maintain greater accuracy than PIR at longer interval durations." [3] In the same 2007 study, momentary estimates differed by less than 5% across intervals from 5 to 120 seconds. [3]
Longer observations give better samples. A longer sample represents the whole session better. [3] In one study they describe (Mudford and Beale, 1990), a low-occurrence behavior needed a 105-minute sample to represent a 2.5-hour session well, while 30 minutes was enough for a high-occurrence behavior. [3]
In a simulation they describe (Devine and colleagues, 2011), 30-second momentary time sampling across a 30-minute session or longer was accurate and sensitive to change. [3] Their advice for now: "practitioners should err on the side of caution and use longer observations when possible" [3], especially for behavior that is rare or varies a lot. [3]
Interval length is often arbitrary. Wirth and colleagues note that many researchers have acknowledged that, in practice, "the selection of interval duration is often arbitrary." [7] They call for practical tools "to help researchers and practitioners determine the magnitude and direction of error they can expect to encounter." [7]
Write it on the sheet. Note the rule, the interval length and the observation length on every data sheet and graph. A percent means little without them.
What a percentage can hide
A change in pattern can look like a change in level. "Measurement also may suggest that the level of behavior has changed when in fact another dimension (e.g., distribution of the behavior over time) has changed." [3] For example, if long, rare episodes of a behavior become short, frequent ones, the total time can stay the same while the pattern changes. [3]
One test with real graphs. Meany-Daboul and colleagues took data sets from five earlier treatment analyses, each recorded continuously and each showing a clear reduction. They redrew the graphs as they would have looked with 10-second momentary time sampling and 10-second partial-interval recording. [3] [6]
Most ratings held. Fiske and Delmolino summarize that the authors found strong correspondence between ratings of functional control across all three methods. That suggests the error from 10-second PIR and MTS "was not sufficient to change ratings of an experimental effect." [3]
The misses leaned one way. Meany-Daboul and colleagues found that "both MTS and PIR were more likely to result in false-negative interpretations than false-positive interpretations for identifying functional relations." [6] A false negative means a real effect was there, but the graph did not show it. They add that "discontinuous measurement systems may be contraindicated for the detection of small treatment effects." [6]
Know the pattern of error. Fiske and Delmolino's bottom line: "practitioners must be cognizant of the characteristic patterns of measurement error that exist with discontinuous measurement." [3]
Measuring yes and no
A sample can miss a short "no." A learner might push the materials away for two seconds between momentary checks, and the interval data would never show it.
Breaux and Smith (2023) note that "A learner will likely use a myriad of topographies to give or withdraw assent" [8], and they describe practice in which "both assent and assent withdrawal are individually defined for the learner, continuously evaluated, and responded to." [8]
So when the main measure is a sample, keep each "no" on its own line. Record it every time it happens, and respond to it right away, not at the next beep.
DRAGEN's rule. We record a learner's "no" as information about the task, the setting or the relationship. It is never "noncompliance" and never a target to reduce. Breaux and Smith write that "any form of the learner's assent withdrawal behaviour should be honored." [8] When a behavior can cause harm, it gets its own definition, its own measure and its own safety plan, kept separate from the learner's ways of saying no.
A made-up example: Student C
Student C is 8 years old. C, the family and the school team agree on a goal: more chances to join play at recess.
- The definition. Joining play means C is taking part in the same game or activity as at least one peer, such as taking a turn, passing a ball or talking about the game. Standing near peers without taking part is not counted.
- The method. Recess lasts 30 minutes, and the aide supervises 12 children, so watching C every second is not possible. The team picks momentary time sampling: a vibrating timer every 30 seconds, 60 checks per recess. At each buzz, the aide looks at C and marks yes or no.
- The starting level. Across the first week, C is joining play at about 6 of 60 checks a day, about 10%.
- The plan. Peers invite C to a game, and C chooses whether to join and which game. A separate line on the sheet records every time C says no to an invitation or walks away, not just at the buzz. Each no is honored. Choosing the swings alone is a choice, not a failure.
- Some weeks later. C is joining play at 27 of 60 checks, 45%. That is an estimate of the share of recess C spends in play, not a count of games.
- The check. On some days a second observer scores the same checks. They agree on 55 of 60, about 92%.
- Reading it honestly. If the "no" line climbs, the team revisits the plan with C and the family before anything else changes.
Common myths
- "Interval data are counts." They are estimates. A marked interval can hold one occurrence or many. [4]
- "All three rules make the same error." Partial-interval tends to run high, whole-interval tends to run low, and momentary time sampling has no set direction. [3]
- "Shorter intervals are always better." Shorter intervals cut systematic error, but they ask more of the observer. In one study, agreement fell to unacceptable levels below 3 seconds. [3]
- "Any interval length will do." Length changes the error, and it is often chosen arbitrarily. [3] [7]
- "A drop in percent always means less behavior." It can also mean the pattern of the behavior over time changed. [3]
- "Discontinuous data are useless." In the five-data-set comparison, ratings made from 10-second PIR and MTS graphs corresponded strongly with the continuous ratings. Small effects are the risk. [3] [6]
Words to know
- Discontinuous measurement: recording a sample of behavior by scoring intervals or moments, not every occurrence.
- Interval: one of the equal blocks of time an observation is split into.
- Partial-interval recording (PIR): mark the interval if the behavior happens at any time in it.
- Whole-interval recording (WIR): mark the interval only if the behavior lasts the whole interval.
- Momentary time sampling (MTS): mark the interval only if the behavior is happening as the interval ends.
- Mixed interval: an interval that the behavior fills only partly.
- Overestimate and underestimate: a result higher or lower than the true level.
- False negative: a real effect that the data fail to show.
- Observation period: the whole time the observer watches, such as one 30-minute recess.
Need a PDF?
Create it here, for free: the DRAGEN app. Use it at your next session or team meeting.
Or print one we already made, also free:
- Interval recording sheet (PDF)
- Which measure? card (PDF)
- Blank graph: percent, 30 sessions (PDF)
- IOA check sheet (PDF)
- Every free PDF
More from DRAGEN
- Lesson: How to count a behavior: frequency, duration, latency and interval recording
- Lesson: How to check your data with interobserver agreement (IOA)
- Lesson: How to read a behavior graph in sixty seconds
- Free app for parents: DRAGENparent
- BCBA care: DRAGENtherapy
- Books: DRAGENbooks
- Talk with Chase: Contact
- Home: dragen.cloud
Related DRAGENwiki pages: measuring behavior · continuous measurement · operational definition · interobserver agreement · accuracy · frequency · duration
What remains unknown
- Fiske and Delmolino write that the longest interval each rule can use before accuracy suffers is hard to specify, and that it likely differs between rules. [3]
- They also write: "Continued research is necessary to further inform the data collection decisions made by practitioners in applied settings." [3]
- The five-data-set comparison used one interval size, 10 seconds. Its authors write: "It is possible that different findings would be obtained when examining treatment procedures aimed at increasing behavior." [6]
- The computer simulation placed target events at random across the observation period. [7] Real sessions may follow other patterns, so its error tables are a guide, not a guarantee.
- The 45% figure comes from one journal's research articles from 1995 to 2005. [3] [5] How often everyday clinics and schools use discontinuous methods today is not known from it.
- This page is general information, not medical or legal advice.
- This page was last checked on November 12, 2026.
Sources
- Behavior Analyst Certification Board. BCBA Test Content Outline (6th ed.). https://www.bacb.com/wp-content/bcba-outline-6thEd/
- Behavior Analyst Certification Board. RBT Test Content Outline (3rd ed.), effective January 1, 2026. https://www.bacb.com/rbt-test-content-outline-3rd-ed
- Fiske, K., & Delmolino, L. (2012). Use of discontinuous methods of data collection in behavioral intervention: Guidelines for practitioners. Behavior Analysis in Practice, 5(2), 77–81. https://doi.org/10.1007/BF03391826 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC3592492/)
- LeBlanc, L. A., Raetz, P. B., Sellers, T. P., & Carr, J. E. (2016). A proposed model for selecting measurement procedures for the assessment and treatment of problem behavior. Behavior Analysis in Practice, 9(1), 77–83. https://doi.org/10.1007/s40617-015-0063-2 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC4788644/)
- Mudford, O. C., Taylor, S. A., & Martin, N. T. (2009). Continuous recording and interobserver agreement algorithms reported in the Journal of Applied Behavior Analysis (1995–2005). Journal of Applied Behavior Analysis, 42(1), 165–169. https://doi.org/10.1901/jaba.2009.42-165 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC2649837/)
- Meany-Daboul, M. G., Roscoe, E. M., Bourret, J. C., & Ahearn, W. H. (2007). A comparison of momentary time sampling and partial-interval recording for evaluating functional relations. Journal of Applied Behavior Analysis, 40(3), 501–514. https://doi.org/10.1901/jaba.2007.40-501 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC1986695/)
- Wirth, O., Slaven, J., & Taylor, M. A. (2014). Interval sampling methods and measurement error: A computer simulation. Journal of Applied Behavior Analysis, 47(1), 83–100. https://doi.org/10.1002/jaba.93 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC4580971/)
- Breaux, C. A., & Smith, K. (2023). Assent in applied behaviour analysis and positive behaviour support: Ethical considerations and practical recommendations. International Journal of Developmental Disabilities, 69(1), 111–121. https://doi.org/10.1080/20473869.2022.2144969 (free full text: https://pmc.ncbi.nlm.nih.gov/articles/PMC9897747/)